Description
Accurate precipitation data are crucial for hydrological modelling and rainwater runoff management. Precipitation variability exists through a wide range of spatial and temporal scales and cannot be observed well using the sparse rain gauge networks. This limitation is further emphasised for urban and mountainous catchments, especially under global warming causing increased frequency of extreme events. Recent advances in remote sensing (RS) techniques help monitoring precipitation over larger areas at more regular resolutions than conventional rain gauge networks. The RS data can be biased mostly due to the indirect estimations prone to multiple error sources and the temporally discrete observations. The wealth of spatiotemporal precipitation data by RS, however, calls for developing data-driven solutions for both the bias correction and hydrological modelling that, in turn, requires new procedures to assure generalization of the existing methods. The present thesis, comprised of a comprehensive summary followed by five appended papers, attempted to evaluate quantitative precipitation estimations (QPE) by state-of-the-art instruments/products for local and regional hydrological applications. Accordingly, two recently installed dual polarimetric doppler X-band weather radars (X-WR) in southern Sweden and multiple Global Precipitation Mission (GPM) products in Iran were studied at the relevant scales for urban hydrology (1–5-min and sub-km) and large water supply river–reservoir system operation (daily-monthly and 0.1°), respectively. The validation against rain gauge observations (Paper I and II) showed a significant dependency of the X-WR and GPM precipitation errors to the radial distance and regional precipitation pattern, respectively. Taking observations from local tipping bucket rain gauges at the 1–30-km ranges as a reference, the apparent problems with a single X-WR was related to the attenuation during heavy rains and overshooting (more obviously at higher elevation angle scans). An internationally bias-corrected GPM product called GPM-IMERG-Final showed a generally good correlation to the synoptic observations of over 300 rain gauges in Iran except for the extreme observations that were much better predicted by GPM-IMERG-Late product during spring, summer, and autumn seasons. To leverage the wealth of spatiotemporally complete, validated precipitation data for the hydrological modelling, two novel data-driven procedures using artificial neural networks (ANNs) were developed. As in Paper III, the formulation of the new ANN input variables, namely, ECOVs and CCOVs, representing the event- and catchment-specific areal precipitation coverage ratios, improved monthly runoff estimations in all the studied sub-catchments of the Karkheh River basin (KRB) in the mountainous semi-arid climate of western Iran. Also, merging the two X-WRs (Paper IV) via an ANN-based QPE relying on the doppler and dual-polarization variables as inputs resulted in improved estimations compared to the single radar and single elevation angle QPE. ANN-assisted estimation of precipitation quantiles also showed better results compared to the merging with an empirically based regression model, especially related to the extreme 5-min data. Finally, Paper V described that impact of human activities such as agricultural developments could rather equally affect on runoff of the KRB, which was considered in Paper III by including some MODIS Terra products as additional inputs.| Period | 2022 Jun 16 |
|---|---|
| Examinee/Supervised person | Hasan Hosseini |
| Degree of Recognition | International |